Commodities don't last forever. There's nobody building Z80A based clusters that I'm aware of. Although they DO still make and sell millions of processors using the Z80 and 8051 instruction sets, there's not much activity in the 8080 or 4004 ballpark these days.
I think it's more that architectural and algorithmic approaches which use masses of general purpose computers tend to have longer lives, because there will always be GP computers. At some level, it's just "recompile for the new CPU" because they have generic architectures, so the algorithms aren't highly optimized for something else. The same can't generally be said of any arbitrary specialized processor, whether it be a CDC7600 vector unit, a FFT box from APS, a T414 from Inmos, or a GPU from Nvidia, etc. A similar situation arises in the DSP world. Your processor of choice might change, discrete logic with those TRW Multiply Accumulate chips, Harvard architecture DSP with a FFT butterfly ALU, or a modern FPGA with multiplier cores. But the basic algorithms and signal flows tend to be pretty consistent. FIR filters are implemented pretty much the same way today as they were back in the 1980s with Schottky TTL and wirewrap, perhaps with a bit more parallelism what with 10s of millions of gates available. Back in the 80s, I did work on a system with a pipelined 1024 point FFT in 16 bit block floating point which now that I count them up is many million gates. If you count the gates in the intermediate storage buffers, (1024 * 2 (I/Q) * 16(bits) is 32kbits of RAM, and at 4 gates/bit for a D-latch, that's already 128kgates. Plus the addressing logic, those nifty 16x16 MAC parts from TRW and "one pass" through the FFT was probably over half a million gates on one (big) board, and there were 9 passes in sequence Designs that leverage peculiarities of a particular problem or platform (e.g. filters with coefficients that are such that a shift and add works) tend to be non-portable Jim Lux -----Original Message----- From: Bret Stouder [mailto:br...@senecadata.com] Sent: Tuesday, April 02, 2013 7:52 AM To: 'Douglas Eadline'; Lux, Jim (337C) Cc: beowulf@beowulf.org Subject: RE: [Beowulf] Roadrunner shutdown A great example of why Xeon or Opteron and now PHI and Kepler make a whole lot more sense than Blue Gene or Origin. :) Commodity off the shelf hardware and software solutions continue to make the world go round............... _______________________________________________ Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
